基于因子图的BDS/IMU列车定位信息融合模型

Information fusion model of BDS/IMU train positioning based on factor graph

精确的列车位置信息是保证列车高速、安全运行的关键。针对北斗卫星导航系统(Beidou Navigation Satellite System,BDS)/惯性测量单元(Inertial Measurement Unit,IMU)列车组合定位模型采用传统信息融合方法误差较大,导致列车定位精度不高的问题,提出基于因子图的BDS/IMU列车定位信息融合模型。采用因子图理论,将BDS和IMU传感器接收到的量测信息抽象为因子节点,状态信息抽象为变量节点,构建BDS/IMU列车定位信息融合因子图模型,当BDS接收有效信号时,只需在因子图的特定时间添加因子节点,实现定位传感器的即插即用。模型中定义了以BDS/IMU列车定位状态信息和量测信息为变量的联合概率密度函数,根据非线性优化理论,通过泰勒展开进行线性化处理,转化为标准最小二乘问题,设计高斯-牛顿迭代的因子图推理算法,求解联合概率分布函数的最大后验估计,计算BDS/IMU列车定位信息的最优估计值,得到列车的精确位置信息。通过模拟数据和实际数据对模型进行验证表明,相比于kalman算法,因子图模型有效降低了列车的位置误差和速度误差,且未出现误差发散现象,有效实现了列车不同定位传感器的非等间隔融合,增强了列车定位的信息融合能力,提高了列车定位的精确性。在实测数据下,列车定位位置均方根误差降至2 m以下,明显小于kalman算法,可为列车的高速可靠运行提供精确的位置信息。

Accurate train position information is the key to ensure the high-speed and safe operation of the train.The traditional information fusion method has larger error in the Beidou Navigation Satellite System(BDS)/Inertial Measurement Unit(IMU) train integrated positioning model,which leads to the low accuracy of train positioning.A BDS/IMU train positioning information fusion model based on factor graph was proposed.Using the factor graph theory,the measurement information received by BDS and IMU sensors was abstracted as factor nodes.The state information was abstracted as variable nodes.The BDS/IMU train positioning information fusion factor graph model was constructed.When BDS received effective signals,it only needed to add factor nodes at a specific time in the factor graph to realize the plug and play of positioning sensors.The joint probability density function with BDS/IMU train positioning status information and measurement information as variables was defined in the model.According to the nonlinear optimization theory,it was linearized through Taylor expansion and transformed into a standard least squares problem.The Gaussian-Newton iterative factor graph inference algorithm is designed to solve the maximum of a posteriori estimation of the joint probability distribution function and calculate the optimal estimation value of BDS/IMU train positioning information.To get the accurate position information of the train.Through the simulated data,the verification of the model shows that the factor graph model effectively reduces the position error and speed error of the train compared with the Kalman algorithm and no error divergence.Thus it effectively realizes the unequal interval fusion of different train positioning sensors,At the same time,it enhances the information fusion ability of train positioning and improves the accuracy of train positioning.Under the measured data,the root means square error of train positioning position is reduced to less than 2 m,which is significantly less than Kalman algorithm.It can provide accurate position information for high-speed and reliable operation of the train.